simulation and modeling of large microwave and millimeter ... · linear network • nonlinear...
TRANSCRIPT
NC STATE UNIVERSITY NC STATE UNIVERSITY
Simulation and Modeling of Large Microwave and Millimeter-Wave Systems, Part 1
Michael Steer with Nikhil Kriplani, Carlos Christofferson and Shivam Priyadarshi
1
Copyright 2009 to 2011 by M. Steer, N. Kriplani and S. Priyadarshi Not to be posted on the web or distributed electronically without permission.
http://www.freeda.org
NC STATE UNIVERSITY
Outline Multi-Physics Simulator (fREEDATM) Minimizes Energy Error (not KCL) Time-domain EM interface Frequency-domain EM (network parameter interface) Thermal, Noise Photonics, Mechanical Molecular Electronics
Code reuse: Uses Trilinos Numerical Libraries from Los Alamos
Simple device modeling Transient Simulation Uncompromised commitment to accuracy Circuit/Field Interaction
2
NC STATE UNIVERSITY
fREEDA
3
Open Source / Open Licensing BSD License (Open to companies to do what they want), well almost there.
NC STATE UNIVERSITY
Essential Concepts for Multi-Physics Modeling • Integration of multi physics problem into the circuit
modeling domain • Object Oriented Approach • Global Modeling
– EM Interconnect Modeling – Digital Macromodeling – Opto Electronic Modeling – Quantum Modeling – Molecular Modeling – Materials Modeling – Bio / Chemical Modeling
• Interfaces
NC STATE UNIVERSITY
Universal Error Concept (Energy Norm)
Universal error concept: AT A TERMINAL Σ FLUX = 0 All Potentials are the SAME
MECHANICAL
THERMAL
ELECTRICAL
FORCE
HEAT CURRENT
I
Flow
POSITION
TEMPERATURE
V
Effort
INERTIAL REFERENCE FRAME
ABSOLUTE ZERO
LOCAL GROUND
LOCAL REFERENCE NODE
For each state variable x there must be a flow and an effort contribution.
i(t) v(t)
5
NC STATE UNIVERSITY
Many Unique Models
• EM Interface (TD and FD) • Electro-Thermal • Dynamic Range • Time Delay • Large Signal Noise • Electro-Optic • Molecular Diode
n+ poly
STI STI
tox
Vg
oxide
L
p type
Jg (tunneling)
V_ref
MOSCAP World’s first implementation
E.G. The tunnel process cannot be expressed as a current-charge-voltage expression as Spice requires. Instead use state variables to implement tunneling correctly.
6
NC STATE UNIVERSITY
Circuit Theory
7
NC STATE UNIVERSITY
Background Reading
NC STATE UNIVERSITY
Background Reading
NC STATE UNIVERSITY
Microstrip Model In a Field Simulator Voltages Are Determined By
Integrating The Electric Field Along a Path In Microstrip Problems the Field is Integrated Over
The Paths Shown to Obtain V The Path of Integration Matters The Dashed Path Yields a Different Value of V2
V 1
V 2
Not the same point electrically.
NC STATE UNIVERSITY
Microstrip Model
Network Model: 11 12
21 22
S SS S
V1 V2 REF REF
REF REF
V2 V1
But a (SPICE) Circuit does not have two reference terminals.
NC STATE UNIVERSITY
Global Reference Node
• Perfect Ground Plane Assumed • (No Retardation Effects) • In A Field Simulator Voltages Are Determined By
Integrating The Electric Field Along a Path • In Microstrip Problems the Field is Integrated
Over The Paths Shown to Obtain • The Path of Integration Matters • The Dashed Path Yields a Totally Different
Value • of
12
Conventional Microwave Circuits (e.g. Microstrip)
V 1
V 2
V and 1
V 2
V 2
Current Microwave Simulators (Linear, Harmonic Balance, Transient) are Based on Nodal Voltage Descriptions To Use Nodal Voltages There Must Be a Common Reference Node Field Simulators Currently Require A Common Reference Ground
NC STATE UNIVERSITY
Local reference node concept • This avoids non-physical connections and therefore is
fundamental for the analysis of spatially distributed circuits as well as for simultaneous thermal-electrical simulations.
V1 V2 REF REF
LOCAL REFERENCE TERMINAL
How to extend this beyond two terminal ports?
Our common view of a port is that it has two terminals
i1
i1
i2
i2
1 2
NC STATE UNIVERSITY
Non- fREEDA fREEDA
Circuit Circuit
14
NC STATE UNIVERSITY
Circuit vs. Network Graph
15
Circuit Graph
Network Graph
Concept of Spice
CIRCUIT GRAPH
NETWORK GRAPH Concept of fREEDA
Must use two circuit graphs to represent a general circuit
NC STATE UNIVERSITY
Network Graph
16
A
N
U
T S
R Q
P
O
F
L
G
H
I B
C J
D
K E
A node a edge a
b
Information is stored in fREEDA as a Network Graph but the difference between a terminal and an element is retained.
c
d
e
f
A
B
C
D
E
F
G
H
I
J
L
M
K
N
O
P
Q R
S
T
U
a b
c
d
e f
h g
g h
NETWORK GRAPH
NC STATE UNIVERSITY
Modeling Concept
17
SPATIALLY DISTRIBUTED CIRCUIT
LINEAR CIRCUIT
NONLINEAR NETWORK
THERMAL NETWORK
LINEAR NETWORK
• Nonlinear Analysis Must be Point of Integration Use SPICE-like & HARMONIC BALANCE analyses
• Use Existing Domain Specific Analyses E.G. Thermal, Field • Behavioral Modeling Is the Key to Using Results of One Simulator In Another
NC STATE UNIVERSITY
Modeling Scope
18
Can handle
11
2 32 31 12 2 3 3
1 1 1
( )( )( ), , ( ), , , ,
( ) ( )( ) ( )( ) , , , , , ,
( ), , ( )
nn
n n
n
dx tdx tx t x t
dt dtd x t d x td x t d x t
y t Fdt dt dt dt
x t x tτ τ
= − −
Where y(t) is either an i(t) or a v(t). Also in any type of analysis we want dy/dx The exact derivatives (w.r.t. time or frequency etc.) we want depend on the type of analysis we are doing (transient, wavelet, harmonic balance). The derivatives needed are calculated using ADOL-C under control of the analysis routines. This is why the same model can be used in any type of analysis.
ADOL-C is one of the many support libraries.
NC STATE UNIVERSITY
fREEDA Unique Feature: Nonlinear devices models based on
state variables: • This provides great flexibility for
the design of new models. All of the analyses are state-variable-based, including a time marching analysis (different integration methods available) and a unique wavelet transient analysis.
• The state variables can be chosen to achieve robust numerical characteristics.
• The calculation of derivatives are free of truncation errors at a small multiple of the run time required to evaluate the original function with little additional memory required. (ADOLC)
CONVENTIONAL DIODE MODEL
STRONG NONLINEARITY v
i
v
x
x
NEW
NEW
MODERATE NONLINEARITY
+
i x
v =
NC STATE UNIVERSITY
Bottom DBR
Top DBR
p-contact
Oxide
Light Output
Active Region n-contact
VCSEL Modeling
Single Mode Rate Equations Carrier density dN(t)/dt =ηi(I(t)-IL(T))/τ – N(t)/τnr – G(T)(N(t)-N0(T))S(t)/(1+εS(t)) Photon density dS(t)/dt = -S(t)/τp + βN(t)/τr + G(T)(N(t)-N0(T))S(t)/(1+εS(t)) Temperature dT(t)/dt = -T(t)/τth + (T0+(I(t)V(t)-P0(t))Rth)/τth
Leakage Current
IL = IL0exp[(-a0+ a1N0 + a2N0T- a3/N0)/T]
Temperature dependence of Gain and Transparency
G(T) = G0(ag0 + ag1T +ag2T2) / (bg0 + bg1T +bg2T2)
N0(T) = Nt0(cn0+cn1T+cn2T2)
Transim Power vs Current
0.00E+00
5.00E-04
1.00E-03
1.50E-03
2.00E-03
2.50E-03
3.00E-03
3.50E-03
0 0.005 0.01 0.015 0.02 0.025 0.03 0.035
Current(amps)
Pow
er(w
atts
)
Power@298K
Power@323K
Power@338KLeakage @298K
Leak
age
(mA
)
0
10
20
30
Transient Optical Output Power
Time(10ns)
drive current
Time(ns)
Wavelength Chirp
DC Optical Power versus Drive Current
Rollover from leakage
NC STATE UNIVERSITY
21
Power and Wavelength degradation due to two components
Electro-Optics
Optical Power Wavelength z=12m z=12m
m
f1=12mm, R=0.04 f2=12mm, R=0.04
Detector
Lens2 Lens1
Vcsel
No feedback
L1 feedback
L1+L2 feedback
No feedback
L1 feedback
L1+L2 feedback
Output power degradation due
to single and double lens feedback
Output wavelength degradation due
to single and double lens feedback
With: Mark Niefeld Ravi Pant Univ. of Arizona
Feedback Results:
NC STATE UNIVERSITY
Time Delays
22
Spice handles only short (< 4 time step) time delays.
TWTA used to validated fREEDA’s ability to handle models with long time delays. Also implemented in many transistor models.
100
101
102
103
104
100
101
102
103
104
Specified delay (ns)
Obs
erve
d de
lay
(ns)
0 2000 4000 6000 8000 10000 12000 14000-1
-0.5
0
0.5
1
Time (ps)
Inpu
t Ter
min
al V
olta
ge (v
olts
)
0 2000 4000 6000 8000 10000 12000 14000-3
-2
-1
0
1
2
3
Time (ps)
Out
put T
erm
inal
Vol
tage
(vol
ts)
fREEDA has no limit
NC STATE UNIVERSITY
Transient Simulation
23
NC STATE UNIVERSITY
24
Choice of Time Step to keep error below a set tolerance.
Conventional approach
Result obtained with linear extrapolation
Result from nonlinear iteration (Trapezoidal / Backward Euler)
Tn+1 Tn Tn-1
Error estimate
Poor estimate of error leads to inappropriate choice of time step
Conventional Time Stepping Algorithm
IDEAL RESULT
The difference between the nonlinear iteration and the straight line extrapolation is taken as an estimate of error. If the error is too large the time step is reduced by a factor of 2. If the error is very low the time step is increased by A factor of 2. This results in excessively short time steps around curves and limits dynamic range.
NC STATE UNIVERSITY
25
fREEDA Time Stepping Algorithm
New approach
Error estimate
Tn+1 Tn Tn-1 Better estimate of error leads to a better choice of time step
Result from
Backward Euler
Result from trapezoidal IDEAL
RESULT
The difference between the Backward Euler and trapezoidal estimates is an incredibly good estimate of error. If the error is too large the time step is reduced by a factor of 2. If the error is very low the time step is increased by A factor of 2. The half way point is taken as the answer.
NC STATE UNIVERSITY
26 -120 -110 -100 -90 -80 -70 -60 -50 -40 -30 -20
-250
-200
-150
-100
-50
0
Input Power at 11GHz(dB)Out
put P
ower
(dB
) at
the
third
ord
er in
term
odul
atio
n pr
oduc
ts
fREEDA-12GHzfREEDA-9GHz ADS-12GHz ADS-9GHz Linear-12GHzMeasured Measured
Fundamental Frequency, f1 = 10GHz, P(f1) = -20dBSecond Tone, f2 = 11GHz
OUTPUT POWER OF TOIAT 9 GHz
OUTPUT POWER OF TOIAT 12 GHz
Spectrum Analyzer Noise Floor
Performs better than Traditional HB simulators
160 dB dynamic range with better than 0.5 dB accuracy.
Excellent agreement with measurements
FEATURES: •High Dynamic Range PHEMT 2-stage Low Noise Amplifier. •8.5 GHz to 14 GHz Frequency Band. •18dB Gain. •+6 V Supply Voltage. •2dB Noise Figure. •Can be used as pre-driver amplifier for phased array radar as well as commercial communications applications.
Two-Tone Test of Dynamic Range (X-Band Amplifier)
LMA411
NC STATE UNIVERSITY
27
0 Tolerance
0.002 0.004 0.006 0.008 0.01
Error
0.01
0.02
0
0.03
0.04
0.05
CONVENTIONAL
NEW
High Dynamic Range Through Error Control
The new approach is a better estimate of error. Better time point selection.
– 160 dBc
NC STATE UNIVERSITY
fREEDA, Transient Simulation and Frequency-Domain Modeling
28
NC STATE UNIVERSITY
Three Milestones for Interfacing EM and Circuits • D. Winkelstein, M. B. Steer, and R. Pomerleau, “Simulation of arbitrary transmission line networks with nonlinear
terminations,” IEEE Trans. on Circuits and Systems, April 1991, pp.418-422. See also, IEEE Trans. on Circuits and Systems, Vol. 38, Oct. 1991.
• Impulse response and convolution • Time and frequency bounded
• C. S. Saunders, J. Hu, C. E. Christoffersen, and M. B. Steer, “Inverse singular value method for enforcing passivity in reduced-order models of distributed structures for transient and steady-state simulation,” IEEE Trans. Microwave Theory and Techniques, April 2011, pp. 837–847.
• S parameters to Foster Model • Passivity, causality
• C. S. Saunders and M. B. Steer, “Passivity enforcement for admittance models of distributed networks using an inverse eigenvalue method,” IEEE Transactions on Microwave Theory and Techniques, In Press.
• y parameters to Foster Model • Passivity, causality • What is needed for circuit simualtpors which are y parameter based
• ALL CONVERTERS NEED SOME HUERITIC KNOWLEDGE • Did the user provide sufficient data
• If not how to extrapolate • What is the basic response (what to emphasize)
• Low pass, Bandpass
29
NC STATE UNIVERSITY
Approaches to Mixing EM Solvers
Frequency-Domain EM Solvers Produce a set of S parameters (or equivalent) Key-Problem is developing a causal/passive/accurate interface for
transient simulation The solution is the Foster Form with Appropriate Fitting Technique
Joining TD EM/ Circuits / FD EM Approach 1 Use fREEDA with FDTD interface and Foster Cannonical Form Interface
Approach 2 Build Foster Model into FDTD
30
NC STATE UNIVERSITY
Workflow 31
EM Model In Frequency Domain Y(f)
Develop Foster N-Port
Time-Domain Circuit Simulation fREEDA
R. Mohan, J. C. Myoung, S. E. Mick, F. P. Hart, K. Chandrasekar, A. C. Cangellaris, P. D. Franzon and M. B. Steer, “Causal reduced-order modeling of distributed structures in a transient circuit simulator,” IEEE Trans. Microwave Theory and Tech, Vol. 52, No. 9, Sept. 2004, pp. 2207 – 2214.
In Review This is the most robust and accurate technique out there. Robustness is not perfect, heuristics required in extraction.
Main issue is in developing model that is 1) Causal 2) Passive 3) Accurate
Use to incorporate FD-EM in transient circuit simulator In xFDTD (not in current scope) but can join
xFDTD through fREEDA to EFD-EM solver
Either implement a Foster’s model directly or synthesize an R, L, C, K subcircuit.
31
NC STATE UNIVERSITY
32
SINGARS SINGARS EM
NC STATE UNIVERSITY
SINCGARS
33
NC STATE UNIVERSITY
Output Waveforms
34
NC STATE UNIVERSITY
Summary
• Most robust circuit simulator technology • Interfaces with TD and FD EM solvers. • Will be supported commercially. • fREEDA papers at
http://people.engr.ncsu.edu/mbs/Publications/mbs_publications.html
35